Skip to content

Comment on Quantitative AI progress needs accurate and transparent evaluation

Comments

My own thoughts on it are that it's entirely crazy that we focus so much on "real world" fixed benchmarks.

I should write an article on it sometime, but I think the incessant focus on data someone collected from the mystical "real world" over well designed synthetic data from a properly understood algorithm is really damaging to proper understanding.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.